/*******************************************************************************

[Last updated: June 4th, 2024]

This script contains the notes printed at the bottom of each Sit-D Table.

*******************************************************************************/




* Table notes for balance, attrition, and unit switch tables -------------------

gl admin_balance_end ///
"\textbf{Notes.} This table examines baseline balance in officer characteristics and officer outcomes in CPD's administrative data during the two years preceding randomization (over January 2018 – January 2020). The first column shows the mean of the control group at baseline; the second column shows the mean of the treatment group at baseline; and the third column presents the difference in means between the treatment and control groups. These estimates are attained by regressing each covariate on the Sit-D indicator, along with stratum (unit x watch) fixed effects. The last column shows the number of observations in these regressions. The last row of the table presents the p-value associated with an F-test of joint significance, from a regression of the Sit-D treatment indicator on all the variables examined in the table, along with stratum fixed effects. *** p \textless 0.01, ** p \textless 0.05, * p \textless 0.1."

gl admin_balance_end_endline ///
"\textbf{Notes.} This table examines baseline balance in officer characteristics and officer outcomes, restricting the sample to officers who completed an endline assessment. The first column shows the mean of the control group at baseline; the second column shows the mean of the treatment group at baseline; and the third column presents the difference in means between the treatment and control groups. These estimates are attained by regressing each covariate on the Sit-D indicator, along with stratum (unit x watch) fixed effects. The last column shows the number of observations in these regressions. The last row of the table presents the p-value associated with an F-test of joint significance, from a regression of the Sit-D treatment indicator on all the variables examined in the table, along with stratum fixed effects. *** p \textless 0.01, ** p \textless 0.05, * p \textless 0.1."

gl admin_attrition_end ///
"\textbf{Notes.} This table examines whether attrition is predicted by treatment status. The first four rows measure attrition out of CPD's administrative data, gauging if the officer is missing in this data source for all months after January 2021 (in the top row); over March 2021-February 2022 (in the second row); over July 2021-February 2022 (in the third row); and over November 2021 – February 2022 (in the bottom row). The bottom row defines attrition as missing from the endline assessment. Each row represents a different regression in which the attrition indicator is regressed on the Sit-D indicator. All regressions include stratum (unit x watch) fixed effects.  Robust standard errors are shown in parentheses. *** p \textless 0.01, ** p \textless 0.05, * p \textless 0.1."

gl admin_unit_switch ///
"\textbf{Notes.} This table examines if the Sit-D treatment affected officer tendencies to switch away from the location in which they were working at the time of randomization. Panel A measures switching away from the unit $\times$ watch; and Panel B from just the unit. Each panel presents three measures: whether the officer ever switched in any post-training month; switched for more than half the post-training months; or switched for all of the post-training months. Each row is one regression. All regressions include stratum fixed effects. Column (1) shows the control means for each outcome.  Column (2) presents the coefficients on the Sit-D indicator. Column (3) shows robust standard errors in parentheses.  Column (4) shows the number of observations in each regression.  *** is significant at the 1\% level, ** is significant at the 5\% level, and * significant at the 10\% level."

gl end_crime_rate ///
"\textbf{Notes.} This table examines how the ratio of Sit-D officers to total officers in each district-watch at the time of randomization affects its crime rate over various periods of time. The dependent variable in Panel A is crimes per 1,000 persons and in Panel B is violent crimes per 1,000 persons. In each panel, a four-month pre-training period (from October 2019-January 2020) and a twelve-month pre-training period (from February 2019 to January 2020) are examined. In addition, a four-month post-training period (from January 2021-April 2021) and a twelve-month post-training period (from January 2021 to December 2021) are also examined. Each row is a separate regression. Four (twelve) monthly observations are included for each district-watch in the four-month (twelve-month) period regressions. All regressions include month fixed effects.  Column (1) shows the coefficient on the Sit-D Ratio. Column (2) presents the standard error, clustered on district-watch. Column (3) presents the p-value and column (4) presents the number of observations. *** is significant at the 1\% level, ** is significant at the 5\% level, and * significant at the 10\% level."
		
* Table notes for Admin tables -------------------------------------------------

capture: program drop write_end
program write_end 
args note_name N1 N2

gl end_key_common "\textbf{Notes.} This table shows the effect of Sit-D training on key field outcomes based on estimating equation \eqref{eqn:Period_ITT}. Each row is a separate regression.  Four monthly post-training observations are included for each officer.  N=`N1'.   All regressions include stratum fixed effects, month fixed effects, and additional officer-level covariates (race, gender, experience; as well as baseline values of discretionary arrests, officer injuries and an index of officer activities, key outcomes from the administrative data measured similarly at baseline and endline). Outcomes are measured per 1,000 officers per month, except officer injuries, which is measured per officer per month. Column (1) shows the control mean for each outcome.  Column (2) presents the coefficient on the Sit-D indicator from estimating equation \eqref{eqn:Period_ITT}. Column (3) shows the standard errors, clustered on officer, and column (4) shows the observed p-value.  Column (5) presents the multiple-inference corrected q-values that adjust for the false discovery rate across outcomes in a family. Uses of non-lethal force and discretionary arrests constitute the Adverse Policing Outcomes Family, and officer injuries and the officer activities index constitute the Officer Safety and Activity Family. *** is significant at the 1\% level , ** is significant at the 5\% level, and * is significant at the 10\% level."
	
gl end_auxiliary ///
"\textbf{Notes.} This table shows the effect of Sit-D training on auxiliary field outcomes based on estimating equation \eqref{eqn:Period_ITT}. Each row is a separate regression.  Four monthly post-training observations are included for each officer.  N=`N1'. All regressions iinclude stratum fixed effects, month fixed effects, and additional officer-level covariates (race, gender, experience; as well as baseline values of discretionary arrests, officer injuries and an index of officer activities, key outcomes from the administrative data measured similarly at baseline and endline). Panel A presents effects on specific use of force measures from the TRRs, and Panel B presents effects on additional outcomes from the TRRs. Outcomes are measured per 1,000 officers per month, except for tactics used in uses of force, which is measured per officer per month. Column (1) shows the control mean for each outcome (blank for mean effect indices).  Column (2) presents the coefficient on the Sit-D indicator from equation \eqref{eqn:Period_ITT}. Column (3) shows the standard errors, clustered on officer, and column (4) shows the observed p-value.  Column (5) presents the multiple-inference corrected q-values that adjust for the false discovery rate across outcomes in a family. Subject injuries, subject allegations of injuries, the hospitalization variables and the tactics variables constitute the Auxiliary TRR family. *** is significant at the 1\% level , ** is significant at the 5\% level, and * is significant at the 10\% level."

gl end_downstream ///
"\textbf{Notes.} This table presents estimates of the Sit-D training on additional outcomes that are downstream from an officers' actions, based on estimating equation \eqref{eqn:Period_ITT}. Each row is a separate regression.  Four monthly post-training observations are included for each officer.  N=`N1'.   All regressions include stratum fixed effects, month fixed effects, and additional officer-level covariates (race, gender, experience; as well as baseline values of discretionary arrests, officer injuries and an index of officer activities, key outcomes from the administrative data measured similarly at baseline and endline). Outcomes are measured per 1,000 officers per month, except for force and abuse related complaints, which is measured per officer per month. Column (1) shows the control mean for each outcome  (blank for mean effect indices).  Column (2) presents the coefficient on the Sit-D indicator from equation \eqref{eqn:Period_ITT}. Column (3) shows the standard errors, clustered on officer, and column (4) shows the observed p-value.  Column (5) presents the multiple-inference corrected q-values that adjust for the false discovery rate across outcomes in a family. The outcomes in this table constitute the Downstream Actions from Officers' Actions Family. *** is significant at the 1\% level , ** is significant at the 5\% level, and * is significant at the 10\% level."
	
gl end_key_3m ///
"\textbf{Notes.} This table shows the effect of Sit-D training on key field outcomes in a hypothetical alternative focal period three months after the training. Each row is a separate regression.  Three monthly post-training observations are included for each officer.  N=`N1'.   All regressions include stratum fixed effects, month fixed effects, and additional officer-level covariates (race, gender, experience; as well as baseline values of discretionary arrests, officer injuries and an index of officer activities, key outcomes from the administrative data measured similarly at baseline and endline). Outcomes are measured per 1,000 officers per month, except officer injuries, which is measured per officer per month. Column (1) shows the control mean for each outcome (blank for mean effect indices).  Column (2) presents the coefficient on the Sit-D indicator from equation \eqref{eqn:Period_ITT}. Column (3) shows the standard errors, clustered on officer, and column (4) shows the observed p-value.  Column (5) presents the multiple-inference corrected q-values that adjust for the false discovery rate across outcomes in a family. Uses of non-lethal force and discretionary arrests constitute the Adverse Policing Outcomes Family, and officer injuries and the officer activities index constitute the Officer Safety and Activity Family.  *** is significant at the 1\% level , ** is significant at the 5\% level, and * is significant at the 10\% level."
	
gl end_key_12m ///
"\textbf{Notes.} This table shows the effect of Sit-D training on key field outcomes 12 months after the training. Each row is a separate regression. Twelve monthly post-training observations are included for each officer. N=`N1'. All regressions include stratum fixed effects, month fixed effects, and additional officer-level covariates (race, gender, experience; as well as baseline values of discretionary arrests, officer injuries and an index of officer activities, key outcomes from the administrative data measured similarly at baseline and endline). Outcomes are measured per 1,000 officers per month, except officer injuries, which is measured per officer per month. Column (1) shows the control mean for each outcome (blank for mean effect indices). Column (2) presents the coefficient on the Sit-D indicator from equation \eqref{eqn:Period_ITT}. Column (3) shows the standard errors, clustered on officer, and column (4) shows the observed p-value.  Column (5) presents the multiple-inference corrected q-values that adjust for the false discovery rate across outcomes in a family. Uses of non-lethal force and discretionary arrests constitute the Adverse Policing Outcomes Family, and officer injuries and the officer activities index constitute the Officer Safety and Activity Family. *** is significant at the 1\% level , ** is significant at the 5\% level, and * is significant at the 10\% level."
	
gl end_key_no_base ///
"\textbf{Notes.} This table presents estimates of the Sit-D training on key field outcomes (shown in each panel). Outcomes are measured per 1,000 officers per month, except officer injuries, which is measured per officer per month. Each row is a different regression. One observation is included for each officer-month. Each outcome is regressed on the interaction of Sit-D with period indicators for Months 1-4, Months 5-8 and Months 9-12 after the training. All regressions also include period indicators, stratum fixed effects and month fixed effects, but do not include any additional covariates. Treatment effects are presented for each four-month period in each panel.  Column (1) denotes the period (i.e., months after training). Column (2) shows the control means for each outcome.  Column (3) presents the coefficients on the interaction of Sit-D with the period indicators.  Column (4) shows the standard errors, clustered on officer, in parentheses. Column (5) shows in square brackets the multiple-inference corrected q-values that adjust for the false discovery rate within each four-month period and across outcomes in a family. Uses of non-lethal force and discretionary arrests constitute the Adverse Policing Outcomes Family, and officer injuries and the officer activities index constitute the Officer Safety and Activity Family.  Column (6) shows the number of observations in each regression.  *** is significant at the 1\% level, ** is significant at the 5\% level, and * is significant at the 10\% level. "
	
gl end_key_het_experience ///
"\textbf{Notes.} This table presents heterogeneous effects of the Sit-D training by officer experience, based on estimating equation \eqref{eqn:Period_Hetero}. Each panel is a separate regression.  Four monthly post-training observations are included for each officer.  N=`N1'.  All outcomes are measured per 1,000 officers per month, except officer injuries, which is measured per officer per month. All regressions include stratum fixed effects, month fixed effects, and additional officer-level covariates (race, gender, experience; as well as baseline values of discretionary arrests, officer injuries and an index of officer activities, key outcomes from the administrative data measured similarly at baseline and endline). Columns (1)-(3) show the coefficient, standard error and p-value for estimates of the Sit-D indicator. Columns (4)-(6) show the coefficient, standard error and p-value for estimates of Sit-D interacted with experience. Standard errors are clustered on officer. *** is significant at the 1\% level, ** is significant at the 5\% level, and * is significant at the 10\% level. "

gl end_key_het_e_race_gender ///
"\textbf{Notes.} This table presents heterogeneous effects of the Sit-D training by officer experience (Panel A), race (Panel B), and gender (Panel C), based on estimating equation \eqref{eqn:Period_Hetero}. Each row is a separate regression. Four monthly post-training observations are included for each officer. N=`N1'. All outcomes are measured per 1,000 officers per month. All regressions include stratum fixed effects, month fixed effects, and additional officer-level covariates (race, gender, experience; as well as baseline values of discretionary arrests, officer injuries and an index of officer activities, key outcomes from the administrative data measured similarly at baseline and endline). Columns (1)-(3) show the coefficient, standard error and p-value for estimates of the Sit-D indicator. Columns (4)-(6) show the coefficient, standard error and p-value for estimates of Sit-D interacted with officer race in Panel A, gender in Panel B, and experience in Panel C. Standard errors are clustered on officer. *** is significant at the 1\% level, ** is significant at the 5\% level, and * is significant at the 10\% level."

gl end_key_het_race_district ///
"\textbf{Notes.} This table presents heterogeneous effects of the Sit-D training by the racial composition of the location in which an officer is employed, based on estimating \eqref{eqn:Period_Hetero}. Each row is a separate regression.  Four monthly post-training observations are included for each officer.  N=`N1'.   All outcomes are measured per 1,000 officers per month, except officer injuries, which is measured per officer per month.  All regressions include stratum fixed effects, month fixed effects, and additional officer-level covariates (the fraction of the district of employment that is Black or Hispanic, race, gender, experience; as well as baseline values of discretionary arrests, officer injuries and an index of officer activities, key outcomes from the administrative data measured similarly at baseline and endline). Columns (1)-(3) show the coefficient, standard error and p-value for estimates of the Sit-D indicator. Columns (4)-(6) show the coefficient, standard error and p-value for estimates of Sit-D interacted with the proportion of the district that is Black or Hispanic. Standard errors are clustered on officer. *** is significant at the 1\% level, ** is significant at the 5\% level, and * is significant at the 10\% level. "

gl end_key_het_race_B_district ///
"\textbf{Notes.} This table presents heterogeneous effects of the Sit-D training by the racial composition of the location in which an officer is employed, based on estimating \eqref{eqn:Period_Hetero}. Each row is a separate regression.  Four monthly post-training observations are included for each officer.  N=`N1'.   All outcomes are measured per 1,000 officers per month, except officer injuries, which is measured per officer per month.  All regressions include stratum fixed effects, month fixed effects, and additional officer-level covariates (the fraction of the district of employment that is Black, race, gender, experience; as well as baseline values of discretionary arrests, officer injuries and an index of officer activities, key outcomes from the administrative data measured similarly at baseline and endline). Columns (1)-(3) show the coefficient, standard error and p-value for estimates of the Sit-D indicator. Columns (4)-(6) show the coefficient, standard error and p-value for estimates of Sit-D interacted with the proportion of the district that is Black. Standard errors are clustered on officer. *** is significant at the 1\% level, ** is significant at the 5\% level, and * is significant at the 10\% level. "

gl end_key_het_race_H_district ///
"\textbf{Notes.} This table presents heterogeneous effects of the Sit-D training by the racial composition of the location in which an officer is employed, based on estimating \eqref{eqn:Period_Hetero}. Each row is a separate regression.  Four monthly post-training observations are included for each officer.  N=`N1'.   All outcomes are measured per 1,000 officers per month, except officer injuries, which is measured per officer per month.  All regressions include stratum fixed effects, month fixed effects, and additional officer-level covariates (the fraction of the district of employment that is Hispanic, race, gender, experience; as well as baseline values of discretionary arrests, officer injuries and an index of officer activities, key outcomes from the administrative data measured similarly at baseline and endline). Columns (1)-(3) show the coefficient, standard error and p-value for estimates of the Sit-D indicator. Columns (4)-(6) show the coefficient, standard error and p-value for estimates of Sit-D interacted with the proportion of the district that is Hispanic. Standard errors are clustered on officer. *** is significant at the 1\% level, ** is significant at the 5\% level, and * is significant at the 10\% level. "

gl end_key_het_race_mB_district ///
"\textbf{Notes.} This table presents heterogeneous effects of the Sit-D training by the racial composition of the location in which an officer is employed, based on estimating \eqref{eqn:Period_Hetero}. Each row is a separate regression.  Four monthly post-training observations are included for each officer.  N=`N1'.   All outcomes are measured per 1,000 officers per month, except officer injuries, which is measured per officer per month.  All regressions include stratum fixed effects, month fixed effects, and additional officer-level covariates (the indication of whether the majority residents in of the district of employment is Black, race, gender, experience; as well as baseline values of discretionary arrests, officer injuries and an index of officer activities, key outcomes from the administrative data measured similarly at baseline and endline). Columns (1)-(3) show the coefficient, standard error and p-value for estimates of the Sit-D indicator. Columns (4)-(6) show the coefficient, standard error and p-value for estimates of Sit-D interacted with the proportion of the district that is Black. Standard errors are clustered on officer. *** is significant at the 1\% level, ** is significant at the 5\% level, and * is significant at the 10\% level. "

gl end_key_het_race_mH_district ///
"\textbf{Notes.} This table presents heterogeneous effects of the Sit-D training by the racial composition of the location in which an officer is employed, based on estimating \eqref{eqn:Period_Hetero}. Each row is a separate regression.  Four monthly post-training observations are included for each officer.  N=`N1'.   All outcomes are measured per 1,000 officers per month, except officer injuries, which is measured per officer per month.  All regressions include stratum fixed effects, month fixed effects, and additional officer-level covariates (the indication of whether the majority residents in of the district of employment is Hispanic, race, gender, experience; as well as baseline values of discretionary arrests, officer injuries and an index of officer activities, key outcomes from the administrative data measured similarly at baseline and endline). Columns (1)-(3) show the coefficient, standard error and p-value for estimates of the Sit-D indicator. Columns (4)-(6) show the coefficient, standard error and p-value for estimates of Sit-D interacted with the proportion of the district that is Hispanic. Standard errors are clustered on officer. *** is significant at the 1\% level, ** is significant at the 5\% level, and * is significant at the 10\% level. "

gl end_key_het_ratio ///
"\textbf{Notes.} This table examines spillover effects of the Sit-D training on key field outcomes. Each panel is a separate regression. Four monthly post-training observations are included for each officer. N=`N1'. Outcomes are measured per 1,000 officers per month, except officer injuries, which is measured per officer per month. Column (1) shows the control mean for each outcome. Column (2)-(4) show the coefficient, standard error and p-value for estimates of  the Sit-D indicator. Column (5)-(7) show the coefficients, standard error and p-value for estimates of the Sit-D indicator interacted with the ratio of treated officers working in a given unit x watch in a given month. Standard errors are clustered on this unit x watch. All regressions also include the un-interacted ratio, stratum fixed effects, month fixed effects, and additional officer-level covariates (experience, race and gender, and baseline values of all dependent variables measured in the same way at baseline and endline). *** is significant at the 1\% level , ** is significant at the 5\% level, and * is significant at the 10\% level."


gl end_spill_over_ANA ///
"\textbf{Notes.} This table examines spillover effects of Sit-D in the focal period (1-4 months after the training). It presents heterogeneous effects by the  ratio of Sit-D officers to total officers in each unit-watch (stratum) at the time of randomization,  on the basis of estimating \eqref{eqn:Spillover}. Each row is a separate regression. Four monthly post-training observations are included for each officer. N=`N1'. All outcomes are measured per 1,000 officers per month, except officer injuries, which is measured per officer per month. All regressions include stratum fixed effects, month fixed effects, and additional officer-level covariates (race, gender, experience; as well as baseline values of discretionary arrests, officer injuries and an index of officer activities, key outcomes from the administrative data measured similarly at baseline and endline). Columns (1)-(3) show the coefficient, standard error and p-value for estimates of the Sit-D indicator. Columns (4)-(6) show the coefficient, standard error and p-value for estimates of Sit-D interacted with the Sit-D ratio. Standard errors are clustered on the unit-watch (stratum). *** is significant at the 1\% level , ** is significant at the 5\% level, and * is significant at the 10\% level."

gl end_spill_over_ANA_extra ///
"\textbf{Notes.} This table examines spillover effects of Sit-D in additional later periods (5-8 months and 9-12 months after the training, in Panels A and B, respectively). It presents heterogeneous effects by the ratio of Sit-D officers to total officers in each unit-watch (stratum) at the time of randomization, on the basis of estimating \eqref{eqn:Spillover}. Each row is a separate regression. Four monthly post-training observations are included for each officer, in each panel. N=`N1' in Panel A and N=`N2' in Panel B. All outcomes are measured per 1,000 officers per month, except officer injuries, which is measured per officer per month. All regressions include stratum fixed effects, month fixed effects, and additional officer-level covariates (race, gender, experience; as well as baseline values of discretionary arrests, officer injuries and an index of officer activities, key outcomes from the administrative data measured similarly at baseline and endline). Columns (1)-(3) show the coefficient, standard error and p-value for estimates of the Sit-D indicator. Columns (4)-(6) show the coefficient, standard error and p-value for estimates of Sit-D interacted with the Sit-D ratio. Standard errors are clustered on the unit-watch (stratum). *** is significant at the 1\% level , ** is significant at the 5\% level, and * is significant at the 10\% level."


gl end_arr_1 ///
"\textbf{Notes.} This table shows heterogeneous effects of the Sit-D training on arrests, for Black subjects and subjects of all other races. Each row is a separate regression, based on estimating equation \eqref{eqn:Period_ITT}. Four monthly post-training observations are included for each officer. N=`N1'. Outcomes are measured per 1,000 officers per month. Discretionary arrests comprise our pre-specified categories; other arrests comprise all other arrests that do not fall under the discretionary arrests variable; and all arrests comprise the sum of discretionary and other arrests, spanning all arrests made in that month. All regressions include stratum fixed effects, month fixed effects, and additional officer-level covariates (race, gender, experience; as well as baseline values of discretionary arrests, officer injuries and an index of officer activities, key outcomes from the administrative data measured similarly at baseline and endline). Column (1) shows the control mean for each outcome. Column (2) presents the coefficient on the Sit-D indicator from estimating equation \eqref{eqn:Period_ITT}. Column (3) shows the standard errors, clustered on officer, and column (4) shows the observed p-value. *** is significant at the 1\% level, ** is significant at the 5\% level, and * is significant at the 10\% level."

gl end_arr_1_z ///
"\textbf{Notes.} This table shows heterogeneous effects of the Sit-D training on arrests, for Black subjects and subjects of all other races. Each row is a separate regression, based on estimating equation \eqref{eqn:Period_ITT}. Four monthly post-training observations are included for each officer. N=`N1'. Outcomes (per officer per month) have been converted to Z-scores. "Discretionary arrests" comprise our pre-specified categories; "other arrests" comprise all other arrests that do not fall under the discretionary arrests variable; and "all arrests" comprise the sum of discretionary and other arrests, spanning all arrests made in that month. All regressions include stratum fixed effects, month fixed effects, and additional officer-level covariates (race, gender, experience; as well as baseline values of discretionary arrests, officer injuries and an index of officer activities, key outcomes from the administrative data measured similarly at baseline and endline). Column (1) shows the control mean for each outcome (blank for mean effect indices). Column (2) presents the coefficient on the Sit-D indicator from estimating equation \eqref{eqn:Period_ITT}. Column (3) shows the standard errors, clustered on officer, and column (4) shows the observed p-value. *** is significant at the 1\% level, ** is significant at the 5\% level, and * is significant at the 10\% level."

gl end_arr_2 ///
"\textbf{Notes.} This table shows heterogenous effects of the Sit-D training on arrests, based on subject race. The race categories are: Black, White, Hispanic and All other races (which include Asian/Pacific Islander and Native American).  Each row is a separate regression, as given by equation \eqref{eqn:Period_ITT}. Four monthly post-training observations are included for each officer. N=`N1'. Outcomes are measured per 1,000 officers per month. Discretionary arrests comprise our pre-specified categories; Other arrests comprise all other arrests that do not fall under the discretionary arrests variable; and All arrests comprise the sum of discretionary and other arrests, spanning all arrests made in that month. There were no Discretionary Arrests in the All other races category in our sample, so the table does not include this regression.  All regressions include stratum fixed effects, month fixed effects, and additional officer-level covariates (race, gender, experience; as well as baseline values of discretionary arrests, officer injuries and an index of officer activities, key outcomes from the administrative data measured similarly at baseline and endline). Column (1) shows the control mean for each outcome. Column (2) presents the coefficient on the Sit-D indicator from estimating equation \eqref{eqn:Period_ITT}. Column (3) shows the standard errors, clustered on officer, and column (4) shows the observed p-value. *** is significant at the 1\% level, ** is significant at the 5\% level, and * is significant at the 10\% level."

gl end_arr_NI ///
"\textbf{Notes.} This table shows the effect of Sit-D training on arrests for Non-index crimes (as defined by \cite{ba_2022}), based on estimating equation \eqref{eqn:Period_ITT}. Each panel is a separate regression. Four monthly post-training observations are included for each officer. N=`N1'. All regressions include stratum fixed effects, month fixed effects, and additional officer-level covariates (race, gender, experience; as well as baseline values of discretionary arrests, officer injuries and an index of officer activities, key outcomes from the administrative data measured similarly at baseline and endline).  The top row presents the sum of all arrests classified as non-index crimes, while remaining rows separately show effects individually on each FBI charge category constituting the sum. Outcomes are measured per 1,000 officers per month. Column (1) shows the control mean for each outcome. Column (2) presents the coefficient on the Sit-D indicator from estimating equation \eqref{eqn:Period_ITT}. Column (3) shows the standard errors, clustered on officer, and column (4) shows the observed p-value. *** is significant at the 1\% level , ** is significant at the 5\% level, and * is significant at the 10\% level."

gl end_arr_OI ///
"\textbf{Notes.} This table shows heterogeneous effects of the Sit-D training on arrests, based on subject race. The race categories are: Black, White, Hispanic and All other races (which include Asian/Pacific Islander and Native American). Each row is a separate regression, based on estimating equation \eqref{eqn:Period_ITT}. Four monthly post-training observations are included for each officer. N=`N1'. Outcomes are measured per 1,000 officers per month. "Discretionary arrests" comprise our pre-specified categories; "other arrests" comprise all other arrests that do not fall under the discretionary arrests variable; and "all arrests" comprise the sum of discretionary and other arrests, spanning all arrests made in that month. There were no discretionary arrests in the all other races category in our sample, so the table does not include this regression. All regressions include stratum fixed effects, month fixed effects, and additional officer-level covariates (race, gender, experience; as well as baseline values of discretionary arrests, officer injuries and an index of officer activities, key outcomes from the administrative data measured similarly at baseline and endline). Column (1) shows the control mean for each outcome. Column (2) presents the coefficient on the Sit-D indicator from estimating equation \eqref{eqn:Period_ITT}. Column (3) shows the standard errors, clustered on officer, and column (4) shows the observed p-value. *** is significant at the 1\% level, ** is significant at the 5\% level, and * is significant at the 10\% level."

gl end_BN_arr_vio_pro_ni ///
"\textbf{Notes.} This table shows heterogeneous effects of the Sit-D training on violent, property and non-index arrests, for Black subjects and subjects of all other races. Each row is a separate regression, based on estimating equation \eqref{eqn:Period_ITT}. Four monthly post-training observations are included for each officer. N=`N1'. Outcomes are measured per 1,000 officers per month. All regressions include stratum fixed effects, month fixed effects, and additional officer-level covariates (race, gender, experience; as well as baseline values of discretionary arrests, officer injuries and an index of officer activities, key outcomes from the administrative data measured similarly at baseline and endline). Column (1) shows the control mean for each outcome. Column (2) presents the coefficient on the Sit-D indicator from estimating equation \eqref{eqn:Period_ITT}. Column (3) shows the standard errors, clustered on officer, and column (4) shows the observed p-value. *** is significant at the 1\% level, ** is significant at the 5\% level, and * is significant at the 10\% level."

gl end_arr_crime ///
"\textbf{Notes.} This table presents heterogeneous effects of the Sit-D training by crime rates of the district in which an officer is employed, based on estimating \eqref{eqn:Period_Hetero}. Panel A considers violent crime rates and Panel B considers overall crime rates, both calculated over 2018-2020, prior to the start of the training. Each row is a separate regression. Four monthly post-training observations are included for each officer. N=`N1'. All outcomes are measured per 1,000 officers per month. All regressions include stratum fixed effects, month fixed effects, and additional officer-level covariates (race, gender, experience; as well as baseline values of discretionary arrests, officer injuries and an index of officer activities, key outcomes from the administrative data measured similarly at baseline and endline). Columns (1)-(3) show the coefficient, standard error and p-value for estimates of the Sit-D indicator.  Columns (4)-(6) show the coefficient, standard error and p-value for estimates of Sit-D interacted with the district's violent or overall crime rate. Standard errors are clustered on officer. *** is significant at the 1\% level, ** is significant at the 5\% level, and * is significant at the 10\% level."

gl end_arr_baseline ///
"\textbf{Notes.} This table presents heterogeneous effects of the Sit-D training by baseline measures of field outcomes,  measured as the sum of these outcomes over the baseline period, comprising the 25 months prior to randomization. The baseline outcomes examined include uses of non-lethal force (Panel A), discretionary arrests (Panel B), total arrests (Panel C), and total complaints (Panel D). The specification is based on estimating \eqref{eqn:Period_Hetero}. Each row is a separate regression. Four monthly post-training observations are included for each officer. N=`N1'. All outcomes are measured per 1,000 officers per month. All regressions include the standard covariate set described in the notes to \autoref{Admin_key_outcomes}. Columns (1)-(3) show the coefficient, standard error and p-value for estimates of the Sit-D indicator. Columns (4)-(6) show the coefficient, standard error and p-value for estimates of Sit-D interacted with the baseline measure examined in each panel. Standard errors are clustered on officer. *** is significant at the 1\% level, ** is significant at the 5\% level, and * is significant at the 10\% level."

gl end_key_outcomes_alternative ///
"\textbf{Notes.} This table shows the effect of Sit-D training on key field outcomes, varying the control set. Each row is a separate regression. Four monthly post-training observations are included for each officer. N=`N1'. All regressions include stratum fixed effects and month fixed effects. Regressions in Panel A include officer-level covariates incorporated by the LASSO double-selection procedure. Regressions in Panel B do not include any additional covariates. Outcomes are measured per 1,000 officers per month, except officer injuries, which is measured per officer per month. Column (1) shows the control mean for each outcome (blank for mean effect indices). Column (2) presents the coefficient on the Sit-D indicator from estimating equation \eqref{eqn:Period_ITT}. Column (3) shows the standard errors, clustered on officer, and column (4) shows the observed p-value. Column (5) presents the multiple-inference corrected q-values that adjust for the false discovery rate across outcomes in a family. Uses of non-lethal force and discretionary arrests constitute the Adverse Policing Outcomes Family, and officer injuries and the officer activities index constitute the Officer Safety and Activity Family. *** is significant at the 1\% level, ** is significant at the 5\% level, and * is significant at the 10\% level."

gl end_alternative_spec ///
"\textbf{Notes.} This table shows the effect of Sit-D training on key field outcomes using a specification in which each control officer is incorporated into the dataset for all the post-training periods represented among all the treated officers in their stratum. Each row is a separate regression. Four monthly post-training observations are included for each treatment officer, but more than four monthly post-training observations are included for each control officer. N=`N1'.  All regressions include stratum fixed effects, month fixed effects, and additional officer-level covariates (race, gender, experience; as well as baseline values of discretionary arrests, officer injuries and an index of officer activities, key outcomes from the administrative data measured similarly at baseline and endline). Outcomes are measured per 1,000 officers per month, except officer injuries, which is measured per officer per month. Column (1) shows the control mean for each outcome (blank for mean effect indices). Column (2) presents the coefficient on the Sit-D indicator from estimating equation \eqref{eqn:Period_ITT}. Column (3) shows the standard errors, clustered on officer, and column (4) shows the observed p-value. Column (5) presents the multiple-inference corrected q-values that adjust for the false discovery rate across outcomes in a family. Uses of non-lethal force and discretionary arrests constitute the Adverse Policing Outcomes Family, and officer injuries and the officer activities index constitute the Officer Safety and Activity Family. *** is significant at the 1\% level , ** is significant at the 5\% level, and * is significant at the 10\% level."

end